Extraction of clinical phenotypes for Alzheimer disease dementia from clinical notes using natural language processing

Objectives

There is much interest in utilizing clinical data for developing prediction models for Alzheimer disease (AD) risk, progression, and outcomes. Existing studies have mostly utilized curated research registries, image analysis, and structured Electronic Health Record (EHR) data. However, much critical information resides in relatively inaccessible unstructured clinical notes within the EHR.

Materials and Methods

We developed a natural language processing (NLP)-based pipeline to extract AD-related clinical phenotypes, documenting strategies for success and assessing the utility of mining unstructured clinical notes. We evaluated the pipeline against gold-standard manual annotations performed by two clinical dementia experts for AD-related clinical phenotypes including medical comorbidities, biomarkers, neurobehavioral test scores, behavioral indicators of cognitive decline, family history, and neuroimaging findings.

Results

Documentation rates for each phenotype varied in the structured versus unstructured EHR. Inter-annotator agreement was high (Cohen’s kappa = 0.72–1) and positively correlated with the NLP-based phenotype extraction pipeline’s performance (average F1-score = 0.65-0.99) for each phenotype.

Discussion

We developed an automated NLP-based pipeline to extract informative phenotypes that may improve the performance of eventual machine-learning predictive models for AD. In the process, we examined documentation practices for each phenotype relevant to the care of AD patients and identified factors for success.

Conclusion

Success of our NLP-based phenotype extraction pipeline depended on domain-specific knowledge and focus on a specific clinical domain instead of maximizing generalizability.

Methods
We developed a natural language processing (NLP)-based pipeline which contains independent NLP modules that target the extraction of ten clinical phenotypes relevant to Alzheimer disease dementia progression. The pipeline was trained on unstructured clinical notes originating from Allscripts TouchWorks associated with AD dementia patient office vsits that occurred between June 1, 2013, to May 31, 2018, extracted from the Washington University in St. Louis Research Data Core (RDC), a repository of patient clinical data from BJC HealthCare and Washington University Physicians. The targeted phenotypes included neurobehavioral test scores (Clinical Dementia Rating and Mini-Mental State Exam) and their corresponding test dates, comorbidities (hypertension and depression), neuroimaging findings (presence of atrophy or infarct), behavioral indicators of dementia (repeating and misplacing), biomarker levels (total and phosphorylated tau protein levels), and family history (whether there was a family history of dementia, and if yes, which family member(s)).

The clinical notes extracted from EHR were in rich text format (RTF) contained within tab-delimited files (TXT) alongside metadata such as the patient medical record number, author, and date authored. These were preprocessed before being analyzed by the NLP-based phenotype extraction pipeline. This entailed converting the TXT files to comma-separated files (CSV), accounting for additional tab, quote, and newline characters present, and stripping the RTF formatting.